I am confused by these posts. On one hand, Eliezer argues for an account of causality in terms of probability, which as we know are subjective degrees of belief. So we should be able to read off whether X thinks A causes B from looking at conditional probabilities in X’s map.
But on the other hand, he suggests (not completely sure this is his view from the article) that the universe is actually made of cause and effect. I would think that the former argument instead suggests causality is “subjectively objective”. Just as with probability, causality is fundamentally an epistemic relation between me and the universe, despite the fact that there can be widespread agreement on whether A causes B. Of course, I can’t avoid cancer by deciding “smoking doesn’t cause cancer”, just as I can’t win the lottery by deciding that my probability of winning it is .9.
For instance, how would an omniscient agent decide if A causes B according Eliezer’s account of Pearl? I don’t think they would be able to, except maybe in cases where they could count frequencies as a substitute for using probabilities.
OK, let’s say you’re looking down at a full printout of a block universe. Every physical fact for all times specified. Then let’s say you do Solomonoff induction on that printout—find the shortest program that will print it out. Then for every physical fact in your printout, you can find the nearest register in your program it was printed out of. And then you can imagine causal surgery—what happens to your program if cosmic rays change that register at that moment in the run. That gives you a way to construe counterfactuals, from which you can get causality.
ETA: There’s still some degrees of freedom in how this gets construed though. Like, what if the printout I’m compressing has all its info time-reversed—it starts out with details about what we’d call the future, then the present, then the past. Then I’d imagine that the shortest program that’d print that out would process everything forward, store it in an accumulator, then run a reversal on that accumulator to print it out, the problem being that the registers printed out from might be downstream from where the value was. It seems like you need some extra magic to be sure of what you mean by “pretend this fact here had gone the other way”
For instance, how would an omniscient agent decide if A causes B according Eliezer’s account of Pearl?
An omniscient agent would have no reason to decide if A causes B, since causality is a tool for predicting the outcomes of interventions, and the omniscient agent already knows what’s going to happen. The concept of “causality” is only useful from a perspective of limited knowledge, much like probability. And the concept of an “intervention” only makes sense in a level of abstraction where free will is apparent.
Pearl addresses this in slide 47 of this lecture. Causality disappears if you consider the entire universe as your object of investigation.
“Actually” isn’t intended in any sense except emphasis and to express that Eliezer’s view is contrary to my expectations (for instance, “I thought it was a worm, but it was actually a small snake”).
Eliezer does seem to be endorsing the statement that “everything is made of causes and effects”, but I am unsure of his exact position. The maximalist interpretation of this would be, “in the correct complete theory of everything, I expect that causation will be basic, one of the things to which other laws are reduced. It will not be the case that causation is explained in terms of laws that make no mention of causation”. This view I strongly disagree with, not least because I generally think something has gone wrong with one’s philosophy if it predicts something about fundamental physics (like Kant’s a priori deduction that the universe is Euclidean).
I suspect this is not Eliezer’s position, though I am unsure because of his “Timeless Physics” post, which I disagree with (as I lean towards four-dimensionalism) but which seems consonant with the above position in that both are consistent with time being non-fundamental. If he means something weaker, though, I don’t know what it is.
Yes, I think Timeless Physics puts you on the right track, and it should be pretty clear that “causality” doesn’t apply so much at the level of comparing possible states of configuration space, aside from perhaps metaphorically to point to which ones are adjacent to which other ones.
Well, if I may take up Bundle’s question on his behalf, Eliezer said, in the article to which these are comments:
“A universe is a connected fabric of causes and effects.”
and
Does the idea that everything is made of causes and effects meaningfully constrain experience? Can you coherently say how reality might look, if our universe did not have the kind of structure that appears in a causal model?
Where I take the second quote to imply an endorsement on Eliezer’s part of the claim ‘everything is made of causes and effects’.
Also, what work is “actually” doing in that sentence?
I don’t know, I was quoting (without making that clear) Bundle’s phrase.
Where I take the second quote to imply an endorsement on Eliezer’s part of the claim ‘everything is made of causes and effects’.
That was from a koan, and a good one at that. How would one perform an experiment to determine whether the universe operates on causes and effects? It suggests there might be something wrong with the conception that the universe is “made of” causes and effects.
Maybe, though given the introduction to the article, I think the koan is about the question ‘what counts as meaningful?‘, not ‘is the universe made of causal relations’.
But, perhaps more interesting than settling the question of whether or not Eliezer thinks the universe is made of causal relations, we can ask “Is the thesis that the universe is made of causal relations inconsistant with Eliezer’s (and your) views on the objectivity of causal relations?′
Given your responses, my dialectical instincts are telling me you think the answer to the above is ‘Yes, they are inconsistant, and it is false that the universe is made of causal relations’. Is that so?
Given your responses, my dialectical instincts are telling me you think the answer to the above is ‘Yes, they are [inconsistent], and it is false that the universe is made of causal relations’. Is that so?
Yes. I thought I actually made that explicit. At least, it’s not “made of” causal relations any more than it’s “made of” probability.
Thanks. I think Eliezer is endorsing the ‘causal relations are fundamental’ reading, and that this apparently conflicts with the idea that causality is the tool of a limited observer. I think he’s likely to see these as reconcilable in some way. That, at any rate, is my prediction.
For instance, how would an omniscient agent decide if A causes B according Eliezer’s account of Pearl? I don’t think they would be able to, except maybe in cases where they could count frequencies as a substitute for using probabilities.
Indeed. Eliezer motivates causal graphs by pointing to limitations in the observer: It’s just not feasible to gather all of the necessary frequencies to do pure Bayesian conditionalization. But that’s saying something about the observer, not about the object that the observer is talking about.
We’re only talking about causal networks because of our own limitations as epistemic agents. It seems a weird leap to say that reality itself is made out of this stuff (causal networks) that we wouldn’t even be thinking about if it weren’t for our own imperfections.
It seems to me that this is the primary thing that we should be working on. If probability is subjective, and causality reduces to probability, then isn’t causality subjective, i.e., a function of background knowledge?
Causality is a useful tool in the map, to describe the territory. The Territory just is. It doesn’t have causality, laws of physics or anything else. Those are just the best tools we have to form accurate maps of it.
This question seems decision-theory complete. If you can reify causal graphs in situations where you’re in no state of uncertainty, then you should be able to reify them to questions like “what is the output of this computation here” and you can properly specify a wins-at-Newcomb’s-problem decision theory.
An omniscient agent could still describe a causal structure over the universe- it would simply be deterministic (which is a special case of a probabilistic causal structure). For instance, consider a being that knew all the worldlines of all particles in the universe. It could deduce a causal structure by re-describing these worldlines as a particular solution to a local differential equation. The key difference between causal vs. acausal descriptions is whether or not they are local.
I am confused by these posts. On one hand, Eliezer argues for an account of causality in terms of probability, which as we know are subjective degrees of belief. So we should be able to read off whether X thinks A causes B from looking at conditional probabilities in X’s map.
But on the other hand, he suggests (not completely sure this is his view from the article) that the universe is actually made of cause and effect. I would think that the former argument instead suggests causality is “subjectively objective”. Just as with probability, causality is fundamentally an epistemic relation between me and the universe, despite the fact that there can be widespread agreement on whether A causes B. Of course, I can’t avoid cancer by deciding “smoking doesn’t cause cancer”, just as I can’t win the lottery by deciding that my probability of winning it is .9.
For instance, how would an omniscient agent decide if A causes B according Eliezer’s account of Pearl? I don’t think they would be able to, except maybe in cases where they could count frequencies as a substitute for using probabilities.
OK, let’s say you’re looking down at a full printout of a block universe. Every physical fact for all times specified. Then let’s say you do Solomonoff induction on that printout—find the shortest program that will print it out. Then for every physical fact in your printout, you can find the nearest register in your program it was printed out of. And then you can imagine causal surgery—what happens to your program if cosmic rays change that register at that moment in the run. That gives you a way to construe counterfactuals, from which you can get causality.
ETA: There’s still some degrees of freedom in how this gets construed though. Like, what if the printout I’m compressing has all its info time-reversed—it starts out with details about what we’d call the future, then the present, then the past. Then I’d imagine that the shortest program that’d print that out would process everything forward, store it in an accumulator, then run a reversal on that accumulator to print it out, the problem being that the registers printed out from might be downstream from where the value was. It seems like you need some extra magic to be sure of what you mean by “pretend this fact here had gone the other way”
An omniscient agent would have no reason to decide if A causes B, since causality is a tool for predicting the outcomes of interventions, and the omniscient agent already knows what’s going to happen. The concept of “causality” is only useful from a perspective of limited knowledge, much like probability. And the concept of an “intervention” only makes sense in a level of abstraction where free will is apparent.
Pearl addresses this in slide 47 of this lecture. Causality disappears if you consider the entire universe as your object of investigation.
I think Bundle’s question is just that, given the above, how can Eliezer also say that the universe is actually made of cause and effect?
Does Eliezer say that the universe is actually made of cause and effect? Also, what work is “actually” doing in that sentence?
“Actually” isn’t intended in any sense except emphasis and to express that Eliezer’s view is contrary to my expectations (for instance, “I thought it was a worm, but it was actually a small snake”).
Eliezer does seem to be endorsing the statement that “everything is made of causes and effects”, but I am unsure of his exact position. The maximalist interpretation of this would be, “in the correct complete theory of everything, I expect that causation will be basic, one of the things to which other laws are reduced. It will not be the case that causation is explained in terms of laws that make no mention of causation”. This view I strongly disagree with, not least because I generally think something has gone wrong with one’s philosophy if it predicts something about fundamental physics (like Kant’s a priori deduction that the universe is Euclidean).
I suspect this is not Eliezer’s position, though I am unsure because of his “Timeless Physics” post, which I disagree with (as I lean towards four-dimensionalism) but which seems consonant with the above position in that both are consistent with time being non-fundamental. If he means something weaker, though, I don’t know what it is.
Yes, I think Timeless Physics puts you on the right track, and it should be pretty clear that “causality” doesn’t apply so much at the level of comparing possible states of configuration space, aside from perhaps metaphorically to point to which ones are adjacent to which other ones.
Well, if I may take up Bundle’s question on his behalf, Eliezer said, in the article to which these are comments:
and
Where I take the second quote to imply an endorsement on Eliezer’s part of the claim ‘everything is made of causes and effects’.
I don’t know, I was quoting (without making that clear) Bundle’s phrase.
That was from a koan, and a good one at that. How would one perform an experiment to determine whether the universe operates on causes and effects? It suggests there might be something wrong with the conception that the universe is “made of” causes and effects.
Maybe, though given the introduction to the article, I think the koan is about the question ‘what counts as meaningful?‘, not ‘is the universe made of causal relations’.
But, perhaps more interesting than settling the question of whether or not Eliezer thinks the universe is made of causal relations, we can ask “Is the thesis that the universe is made of causal relations inconsistant with Eliezer’s (and your) views on the objectivity of causal relations?′
Given your responses, my dialectical instincts are telling me you think the answer to the above is ‘Yes, they are inconsistant, and it is false that the universe is made of causal relations’. Is that so?
Yes. I thought I actually made that explicit. At least, it’s not “made of” causal relations any more than it’s “made of” probability.
Thanks. I think Eliezer is endorsing the ‘causal relations are fundamental’ reading, and that this apparently conflicts with the idea that causality is the tool of a limited observer. I think he’s likely to see these as reconcilable in some way. That, at any rate, is my prediction.
My prediction is that he never noticed this problem before.
My prediction is that he will not respond to this line of criticism within the next three days.
Indeed. Eliezer motivates causal graphs by pointing to limitations in the observer: It’s just not feasible to gather all of the necessary frequencies to do pure Bayesian conditionalization. But that’s saying something about the observer, not about the object that the observer is talking about.
We’re only talking about causal networks because of our own limitations as epistemic agents. It seems a weird leap to say that reality itself is made out of this stuff (causal networks) that we wouldn’t even be thinking about if it weren’t for our own imperfections.
It seems to me that this is the primary thing that we should be working on. If probability is subjective, and causality reduces to probability, then isn’t causality subjective, i.e., a function of background knowledge?
This seems not in the least contentious, if you’re talking about the map of causality.
The question is whether causality exists in the territory at all.
Causality is a useful tool in the map, to describe the territory. The Territory just is. It doesn’t have causality, laws of physics or anything else. Those are just the best tools we have to form accurate maps of it.
How do you know that?
Thanks for succinctly articulating what was bothering me about this post. Can’t upvote this enough.
This question seems decision-theory complete. If you can reify causal graphs in situations where you’re in no state of uncertainty, then you should be able to reify them to questions like “what is the output of this computation here” and you can properly specify a wins-at-Newcomb’s-problem decision theory.
An omniscient agent could still describe a causal structure over the universe- it would simply be deterministic (which is a special case of a probabilistic causal structure). For instance, consider a being that knew all the worldlines of all particles in the universe. It could deduce a causal structure by re-describing these worldlines as a particular solution to a local differential equation. The key difference between causal vs. acausal descriptions is whether or not they are local.